On X
Posts about software and AI: building products, programming and the AI tools I use.
131 posts
If Fable is already almost unusable even when you’re paying €100/€200 a month, I’m scared to see what happens when the extended limits period ends. Anthropic needs compute now. Well, it’s needed it for months, but now people are realizing just how much you can get done with Codex and even Grok.
2 good places to get ideas for skills: → aihero.dev/skills → github.com/cursor/plugins/tree/main/psta… My favorites are: /unslop /bro /grill-with-docs and I'm starting to get the hang of /figure-it-out
⭐️ I made a mini tool that's 100% free It turns every sentence in your text into a dot, positioned according to its word count. If the dots pile up in a column, all your sentences are the same length and your writing sounds flat 😴
It’s sooo easy to tell which apps have been vibecoded. So easy. There are details that scream it right in your face from minute one. Little things… but they immediately give you away: 🎨 The design. It’s not that it’s ugly. It’s that it’s the default design. You can almost guess which model has
I think people aren't adapting quickly enough to improvements in AI. Agents' capabilities increase considerably every month, and layers of abstraction should increase too.
Last week I was playing around with uncensored Qwen locally. I asked it for “effective life hacks that aren’t ethical.” It pisses me off, because I’ve seen (and suffered) plenty of them. Still, I found its last thought pretty interesting:
Another interesting concept. I asked it: what’s the most extreme and important question I should ask an uncensored LLM? And its answer was:
of course, my next question was “what am I really asking?” A slightly less interesting result, but I like the abstraction it makes: “your last two questions were about leverage.”
To this day, I still can't understand how there are so many software developers in Spain who do nothing. I know a lot of people who get paid to do practically nothing. And the company doesn't get rid of them. A complete waste. For years.
My new project 👇 I’ve wanted to build a tool for investors for a while, but I didn’t want to make the usual thing. I didn’t want to build another portfolio management tool or another TIKR redesign. I wanted to make something a bit more original 👀
Worth noting, I started playing around with local LLMs for 2 reasons: 1. I’d burned through my Codex limits 2. Uncensored Qwen was available We’re at a point where local AI can be really, really useful — and can complement tasks from the main providers.
It happened to me with a couple of candidates last year. What pissed me off wasn’t that they did it, but that they did it badly, denied it when you asked, and took you for an idiot. One guy in particular, his screen would freeze every time he switched over to ChatGPT and the dude didn’t t.co/53bTnd8Zqa
